double EigenSolverPoissonImageEditing::solve(const NamedParameters& solverParameters, const NamedParameters& problemParameters, bool profileSolve, std::vector<SolverIteration>& iters)
{
int numUnknowns = 0;
std::unordered_map<vec2i, int, vec2iHash> pixelLocationsToIndex;
std::vector<vec2i> pixelLocations;
size_t pixelCount = m_dims[0] * m_dims[1];
std::vector<float4> h_unknownFloat(pixelCount);
std::vector<float4> h_target(pixelCount);
std::vector<float> h_mask(pixelCount);
findAndCopyArrayToCPU("X", h_unknownFloat, problemParameters);
findAndCopyArrayToCPU("T", h_target, problemParameters);
findAndCopyArrayToCPU("M", h_mask, problemParameters);
for (int y = 0; y < (int)m_dims[1]; ++y) {
for (int x = 0; x < (int)m_dims[0]; ++x) {
if (h_mask[y*m_dims[0] + x] == 0.0f) {
++numUnknowns;
vec2i p(x, y);
pixelLocationsToIndex[p] =(int)pixelLocations.size();
pixelLocations.push_back(p);
}
}
}
printf("# Unknowns: %d\n", numUnknowns);
int numResiduals = (int)pixelLocations.size() * 4;
Eigen::VectorXf x_r(numUnknowns), b_r(numResiduals);
Eigen::VectorXf x_g(numUnknowns), b_g(numResiduals);
Eigen::VectorXf x_b(numUnknowns), b_b(numResiduals);
Eigen::VectorXf x_a(numUnknowns), b_a(numResiduals);
b_r.setZero();
b_g.setZero();
b_b.setZero();
b_a.setZero();
for (int i = 0; i < pixelLocations.size(); ++i) {
vec2i p = pixelLocations[i];
float4 color = sampleImage(h_unknownFloat.data(), p, m_dims[0]);
x_r[i] = color.x;
//printf("%f\n", color.x);
x_g[i] = color.y;
x_b[i] = color.z;
x_a[i] = color.w;
}
SpMatrixf A(numResiduals, numUnknowns);
A.setZero();
printf("Constructing Matrix\n");
std::vector<Tripf> entriesA;
std::vector<vec2i> offsets;
offsets.push_back(vec2i(-1, 0));
offsets.push_back(vec2i(1, 0));
offsets.push_back(vec2i(0, -1));
offsets.push_back(vec2i(0, 1));
for (int i = 0; i < pixelLocations.size(); ++i) {
vec2i p = pixelLocations[i];
int numInternalNeighbors = 0;
float4 g_p = sampleImage(h_target.data(), p, m_dims[0]);
int j = 0;
for (vec2i off : offsets) {
vec2i q = p + off;
if (q.x >= 0 && q.y >= 0 && q.x < (int)m_dims[0] && q.y < (int)m_dims[1]) {
auto it = pixelLocationsToIndex.find(q);
int row = 4 * i + j;
if (it == pixelLocationsToIndex.end()) {
float4 f_q = sampleImage(h_unknownFloat.data(), q, m_dims[0]);
b_r[row] += f_q.x;
b_g[row] += f_q.y;
b_b[row] += f_q.z;
b_a[row] += f_q.w;
}
else {
entriesA.push_back(Tripf(row, it->second, -1.0f));
}
entriesA.push_back(Tripf(row, i, 1.0f));
float4 g_q = sampleImage(h_target.data(), q, m_dims[0]);
b_r[row] += (g_p.x - g_q.x);
b_g[row] += (g_p.y - g_q.y);
b_b[row] += (g_p.z - g_q.z);
b_a[row] += (g_p.w - g_q.w);
}
++j;
}
}
printf("Entries Set\n");
A.setFromTriplets(entriesA.begin(), entriesA.end());
printf("Sparse Matrix Constructed\n");
A.makeCompressed();
printf("Matrix Compressed\n");
{
float totalCost = 0.0f;
float cost_r = (A*x_r - b_r).squaredNorm();
float cost_g = (A*x_g - b_g).squaredNorm();
float cost_b = (A*x_b - b_b).squaredNorm();
float cost_a = (A*x_a - b_a).squaredNorm();
totalCost = cost_r + cost_g + cost_b + cost_a;
printf("Initial Cost: %f : (%f, %f, %f, %f)\n", totalCost, cost_r, cost_g, cost_b, cost_a);
}
AxEqBSolver solver;
solver.setMaxIterations(97);
printf("Solvers Initialized\n");
clock_t start = clock(), diff;
solver.compute(A);
//printf("solver.compute(A)\n");
solveAxEqb(solver, b_r, x_r);
//printf("Red solve done\n");
solveAxEqb(solver, b_g, x_g);
//printf("Green solve done\n");
solveAxEqb(solver, b_b, x_b);
//printf("Blue solve done\n");
solveAxEqb(solver, b_a, x_a);
diff = clock() - start;
printf("Time taken %f ms\n", diff*1000.0 / double(CLOCKS_PER_SEC));
float totalCost = 0.0f;
float cost_r = (A*x_r - b_r).squaredNorm();
float cost_g = (A*x_g - b_g).squaredNorm();
float cost_b = (A*x_b - b_b).squaredNorm();
float cost_a = (A*x_a - b_a).squaredNorm();
totalCost = cost_r + cost_g + cost_b + cost_a;
printf("Final Cost: %f : (%f, %f, %f, %f)\n", totalCost, cost_r, cost_g, cost_b, cost_a);
for (int i = 0; i < pixelLocations.size(); ++i) {
setPixel(h_unknownFloat.data(), pixelLocations[i], m_dims[0], x_r[i], x_g[i], x_b[i]);
}
findAndCopyToArrayFromCPU("X", h_unknownFloat, problemParameters);;
return (double)totalCost;
}
bool yee_compare(CompareArgs &args)
{
if ((args.image_a_->get_width() != args.image_b_->get_width()) or
(args.image_a_->get_height() != args.image_b_->get_height()))
{
args.error_string_ = "Image dimensions do not match\n";
return false;
}
const auto w = args.image_a_->get_width();
const auto h = args.image_a_->get_height();
const auto dim = w * h;
auto identical = true;
for (auto i = 0u; i < dim; i++)
{
if (args.image_a_->get(i) != args.image_b_->get(i))
{
identical = false;
break;
}
}
if (identical)
{
args.error_string_ = "Images are binary identical\n";
return true;
}
// Assuming colorspaces are in Adobe RGB (1998) convert to XYZ.
std::vector<float> a_lum(dim);
std::vector<float> b_lum(dim);
std::vector<float> a_a(dim);
std::vector<float> b_a(dim);
std::vector<float> a_b(dim);
std::vector<float> b_b(dim);
if (args.verbose_)
{
std::cout << "Converting RGB to XYZ\n";
}
const auto gamma = args.gamma_;
const auto luminance = args.luminance_;
#pragma omp parallel for shared(args, a_lum, b_lum, a_a, a_b, b_a, b_b)
for (auto y = 0; y < static_cast<ptrdiff_t>(h); y++)
{
for (auto x = 0u; x < w; x++)
{
const auto i = x + y * w;
const auto a_color_r = powf(args.image_a_->get_red(i) / 255.0f,
gamma);
const auto a_color_g = powf(args.image_a_->get_green(i) / 255.0f,
gamma);
const auto a_color_b = powf(args.image_a_->get_blue(i) / 255.0f,
gamma);
float a_x;
float a_y;
float a_z;
adobe_rgb_to_xyz(a_color_r, a_color_g, a_color_b, a_x, a_y, a_z);
float l;
xyz_to_lab(a_x, a_y, a_z, l, a_a[i], a_b[i]);
const auto b_color_r = powf(args.image_b_->get_red(i) / 255.0f,
gamma);
const auto b_color_g = powf(args.image_b_->get_green(i) / 255.0f,
gamma);
const auto b_color_b = powf(args.image_b_->get_blue(i) / 255.0f,
gamma);
float b_x;
float b_y;
float b_z;
adobe_rgb_to_xyz(b_color_r, b_color_g, b_color_b, b_x, b_y, b_z);
xyz_to_lab(b_x, b_y, b_z, l, b_a[i], b_b[i]);
a_lum[i] = a_y * luminance;
b_lum[i] = b_y * luminance;
}
}
if (args.verbose_)
{
std::cout << "Constructing Laplacian Pyramids\n";
}
const LPyramid la(a_lum, w, h);
const LPyramid lb(b_lum, w, h);
const auto num_one_degree_pixels =
to_degrees(2 *
std::tan(args.field_of_view_ * to_radians(.5f)));
const auto pixels_per_degree = w / num_one_degree_pixels;
if (args.verbose_)
{
std::cout << "Performing test\n";
}
const auto adaptation_level = adaptation(num_one_degree_pixels);
float cpd[MAX_PYR_LEVELS];
cpd[0] = 0.5f * pixels_per_degree;
for (auto i = 1u; i < MAX_PYR_LEVELS; i++)
{
cpd[i] = 0.5f * cpd[i - 1];
}
const auto csf_max = csf(3.248f, 100.0f);
static_assert(MAX_PYR_LEVELS > 2,
"MAX_PYR_LEVELS must be greater than 2");
float f_freq[MAX_PYR_LEVELS - 2];
for (auto i = 0u; i < MAX_PYR_LEVELS - 2; i++)
{
f_freq[i] = csf_max / csf(cpd[i], 100.0f);
}
auto pixels_failed = 0u;
auto error_sum = 0.;
#pragma omp parallel for reduction(+ : pixels_failed, error_sum) \
shared(args, a_a, a_b, b_a, b_b, cpd, f_freq)
for (auto y = 0; y < static_cast<ptrdiff_t>(h); y++)
{
for (auto x = 0u; x < w; x++)
{
const auto index = y * w + x;
const auto adapt = std::max((la.get_value(x, y, adaptation_level) +
lb.get_value(x, y, adaptation_level)) * 0.5f,
1e-5f);
auto sum_contrast = 0.f;
auto factor = 0.f;
for (auto i = 0u; i < MAX_PYR_LEVELS - 2; i++)
{
const auto n1 =
fabsf(la.get_value(x, y, i) - la.get_value(x, y, i + 1));
const auto n2 =
fabsf(lb.get_value(x, y, i) - lb.get_value(x, y, i + 1));
const auto numerator = std::max(n1, n2);
const auto d1 = fabsf(la.get_value(x, y, i + 2));
const auto d2 = fabsf(lb.get_value(x, y, i + 2));
const auto denominator = std::max(std::max(d1, d2), 1e-5f);
const auto contrast = numerator / denominator;
const auto f_mask = mask(contrast * csf(cpd[i], adapt));
factor += contrast * f_freq[i] * f_mask;
sum_contrast += contrast;
}
sum_contrast = std::max(sum_contrast, 1e-5f);
factor /= sum_contrast;
factor = std::min(std::max(factor, 1.f), 10.f);
const auto delta =
fabsf(la.get_value(x, y, 0) - lb.get_value(x, y, 0));
error_sum += delta;
auto pass = true;
// pure luminance test
if (delta > factor * tvi(adapt))
{
pass = false;
}
if (not args.luminance_only_)
{
// CIE delta E test with modifications
auto color_scale = args.color_factor_;
// ramp down the color test in scotopic regions
if (adapt < 10.0f)
{
// Don't do color test at all.
color_scale = 0.0;
}
const auto da = a_a[index] - b_a[index];
const auto db = a_b[index] - b_b[index];
const auto delta_e = (da * da + db * db) * color_scale;
error_sum += delta_e;
if (delta_e > factor)
{
pass = false;
}
}
if (not pass)
{
pixels_failed++;
if (args.image_difference_)
{
args.image_difference_->set(255, 0, 0, 255, index);
}
}
else
{
if (args.image_difference_)
{
args.image_difference_->set(0, 0, 0, 255, index);
}
}
}
}
const auto error_sum_buff =
std::to_string(error_sum) + " error sum\n";
const auto different =
std::to_string(pixels_failed) + " pixels are different\n";
// Always output image difference if requested.
if (args.image_difference_)
{
args.image_difference_->write_to_file(args.image_difference_->get_name());
args.error_string_ += "Wrote difference image to ";
args.error_string_ += args.image_difference_->get_name();
args.error_string_ += "\n";
}
if (pixels_failed < args.threshold_pixels_)
{
args.error_string_ = "Images are perceptually indistinguishable\n";
args.error_string_ += different;
return true;
}
args.error_string_ = "Images are visibly different\n";
args.error_string_ += different;
if (args.sum_errors_)
{
args.error_string_ += error_sum_buff;
}
return false;
}